National Archives issues guidance clarifying when AI outputs qualify as federal records

The National Archives and Records Administration (NARA) issued guidance clarifying that the mere use of artificial intelligence platforms does not automatically result in the creation of federal records. According to the Aug. 21 memo, whether AI-related materials qualify as federal records depends on how agencies use the inputs and outputs, such as whether the agency relies on them in decision-making, uses them to conduct official business, circulates the material, or incorporates it into an agency system.

NARA emphasized that recorded information does not become a federal record simply because it is made or received by a federal agency. For example, an employee using an AI platform for preliminary research without incorporating the output into their work or sharing it would likely not create a federal record, even if search logs are retained. Conversely, if an agency captures and uses an audit trail from an AI system for an investigation, that audit trail could become a federal record.

The memo also notes that outputs from commercial AI programs such as Gemini and ChatGPT are unlikely to be considered federal records on their own, though agencies retain responsibility for determining whether specific AI materials or outputs meet recordkeeping requirements under the Federal Records Act.

NARA recommends that agencies incorporate federal records management requirements into their internal AI policies and adopt formal AI policies in collaboration with legal, IT, and other stakeholders. The agency acknowledged there is no universal approach, emphasizing that policies should be tailored to each agency’s circumstances.